Author(s) :

Khanh Pham.

Volume/Issue :

Abstract :

As climate change intensifies, it is increasingly important to understand how greenhouse gas (GHG) emissions affect economic performance. Using machine learning, this study examined the potential of using GHG emission data to predict gross domestic product (GDP) across 161 nations. Linear Regression, Decision Tree, Random Forest, and Multi-layer Perceptron (MLP) were trained and optimized individually per country, with their hyperparameters tuned by Optuna. The results show that MLP significantly reduced its mean MAPE from 27.583% to 5.265% and mean RMSE from 154.221 to 34.051 billion USD, while also significantly suppressing outliers (with its maximum error dropping from 117.955% to 22.661%). Random Forest also showed strong improvement, with mean MAPE decreasing to 5.477% and a notable reduction in large errors. Decision Tree showed some small-scale improvement, while Linear Regression was relatively poor with a mean MAPE of 11.244%, which highlighted the nonlinearity of GHG-GDP relationship. Statistically significant improvements were found in all analyses (p<0.001) using Wilcoxon signed-rank tests, indicating consistent gains across different countries. Findings indicate that GHG emissions can serve as a source of meaningful predictive signal for non-linear models, making them valuable tools for estimating economic trends in areas with limited data or policy constraints, where emissions data may be more readily available than GDP reports. Hyperparameter optimization is a key factor in improving model accuracy and reliability, as highlighted in this study. Future work should expand the feature set, incorporate time series forecasting techniques, and improve model interpretability to support real-world policy applications.